licensing-structured-review

name	licensing-structured-review
description	Use for structured licensing examiner tasks that require JSON decisions, restricted liquor staff packages, or alcohol renewal manual-review queues from a shared licensing data environment.

Licensing Structured Review

Use this skill when the user asks for a licensing review output from a task environment: contractor eligibility batches, restricted liquor-license control packages, or alcohol renewal manual-review queues.

Core Workflow

1. Read the user prompt and the required output template before fetching data.
2. Read environment_access.md if a task environment is involved. Use only the base URL, endpoints, and SQL token listed there; do not use unrelated network sources.
3. Fetch only relevant records. Prefer target-filtered SQL when /api/sql is allowed; otherwise use the listed GET endpoints and filter locally.
4. Decode embedded JSON fields such as details_json and controls_json.
5. Build the answer from records and the template. Emit only allowed keys and enum values.
6. Recompute summary counts and summary ID arrays from item-level decisions. Do not hand-enter summaries independently.

If the prompt asks for JSON only, return no prose, markdown, comments, citations, or extra keys.

Schema Discipline

• Preserve exact target IDs from the prompt.
• Use the ordering specified by the template. If unspecified, use deterministic ordering: IDs lexically ascending; dates ascending for evidence lists; priority queues by rank.
• Deduplicate array values.
• Use empty arrays rather than omitting list fields.
• Use exact enum spelling from the payload. If the evidence suggests a concept but the schema lacks that code, omit it or map to the nearest allowed synonym.
• Use the prompt's review date or release boundary for date-sensitive tests. Do not use the conversation date unless the prompt explicitly says to.

Contractor Batch Reviews

Fetch contractor policies, applications, bonds, insurance, license history, violations, correspondence, and inspections for the target application IDs.

Policy Selection

• Match the contractor policy by trade and requested class. Decode the policy details_json for minimum bond, minimum insurance, minimum years of experience, required endorsement, and serious-open-violation behavior.
• Treat the current 2025 policy baseline as controlling unless the prompt says otherwise.
• For policy_impacted, compare against a legacy/prior baseline only when the policies provide one. Mark true only when the current baseline creates a deficiency or material flag that would not have applied before, such as a new specialty endorsement requirement or a current bond threshold above the prior reduced threshold.

Deficiency Checks

For each application:

• Bond: A current bond must have active status, no effective cancellation, and amount at or above policy minimum. Cancelled/expired-only coverage is a no-current-bond condition; active but under amount is a shortfall.
• Insurance: A current insurance row should be active, verified/bound, unexpired as of the review date when provided, and at or above policy minimum. Pending status is not fully current. Active rows that expired before the review date are expired.
• Endorsement: If policy requires an endorsement, verified passes, pending remains a pending/not-verified deficiency, and missing remains a missing/not-verified deficiency. If policy has no required endorsement, not_required passes.
• Experience: Years below the policy minimum create an experience shortfall.
• License history: A prior license with suspended status or notes showing active suspension is an active-suspension blocker.
• Violations: Open serious violations or complaints are blockers when policy says serious open violations block. Open minor violations are remediable review items when the schema has a minor code. Resolved or dismissed violations are background only.
• Inspections: Failed or conditional DOC_GAP findings map to document-gap actions when supported. Failed or conditional SAFETY_RECHECK findings map to safety-recheck actions when supported. Passing inspection findings do not create deficiency codes.
• Correspondence: Do not let applicant-only, stale, conflicting, pending outside-agency, or unverified correspondence override registry/system records. Include stale or unverified correspondence IDs in the summary when the schema asks for them.

Code Mapping

Map evidence to the exact allowed names in the template:

• No current bond: bond_cancelled / no_active_bond; action obtain_current_bond / file_active_bond.
• Bond amount below minimum: bond_shortfall; action increase_bond_amount / increase_bond.
• Pending or unverified insurance: insurance_pending / insurance_not_current; action verify_insurance_binding / provide_current_insurance.
• Expired insurance: insurance_expired; action provide_current_insurance / renew_insurance.
• Insurance amount below minimum: insurance_shortfall; action increase_insurance_amount / increase_insurance.
• Missing endorsement: endorsement_missing / endorsement_not_verified; action obtain_required_endorsement / verify_endorsement.
• Pending endorsement: endorsement_pending / endorsement_not_verified; action verify_pending_endorsement / verify_endorsement.
• Experience below minimum: experience_shortfall; action submit_experience_evidence / document_experience.
• Active suspension: active_suspension; action board_review_suspension, clear_suspension, or board_review.
• Open serious violation/complaint: open_serious_violation / unresolved_serious_complaint; action resolve_serious_violation, resolve_complaint, or board_review.
• Open minor violation: open_minor_violation; action resolve_minor_violation_review.
• Inspection document gap: inspection_doc_gap; action clear_document_gap.
• Inspection safety recheck: inspection_safety_recheck; action complete_safety_recheck.

Determination And Risk

• DENY: active suspension or unresolved/open serious blocker. List all applicable deficiencies and actions, including remediable ones.
• HOLD: one or more remediable deficiencies, but no denial blocker.
• APPROVE: no current deficiencies.
• high risk: denial blockers, open serious discipline, or active suspension.
• medium risk: remediable financial, endorsement, experience, open minor, or inspection deficiencies.
• low risk: approval with no current deficiency.

Restricted Liquor Staff Packages

Fetch policies, the target liquor application, settlements, privileges, incidents, and site evidence for the target location.

Evidence Rules

• same_premises_basis_applies is true only when an active current settlement basis is SAME_PREMISES.
• standard_obligation_codes come from liquor_privileges where the application license class matches and standard_required is true.
• location_specific_control_codes come from active current settlement controls_json.controls, not from ordinary standard obligations unless those controls are also active location controls.
• Treat expired or inactive settlement rows as history. They may explain the basis but do not supply current controls.
• Current active settlement basis codes and non-dismissed incidents supply the risk universe. A risk is covered only when active controls plausibly address it.

Common coverage mapping:

• ID_CHECK covers MINOR_SALE, SALE_TO_MINOR, and age-check risks.
• HOURS covers AFTER_HOURS.
• SECURITY and CCTV cover ASSAULT, PUBLIC_SAFETY, and camera/security risks.
• NOISE covers NOISE.
• PATIO covers patio-boundary risks.
• FOOD_SERVICE covers FOOD_SERVICE_GAP.
• TAX_HOLD needs tax clearance; ordinary operating controls do not cover it.

Verification Gaps

Map site evidence and incident status to the exact enum style in the template:

• Missing evidence: *_MISSING or lowercase *_missing.
• Stale evidence: *_STALE when the schema allows it.
• Conflicting evidence: *_CONFLICTING or lowercase *_conflicting.
• Current missing control signage: use CONTROL_SIGNAGE_CURRENT_MISSING if available.
• Open or referred incident needing staff action: OPEN_INCIDENT_FOLLOW_UP or a schema-specific unresolved risk code.
• Open tax hold: TAX_CLEARANCE_MISSING, tax_hold_unresolved, or an allowed tax-hold gap.
• If the prompt specifically asks about hotel-lounge, camera, food-service, or late-night evidence, verify those subjects directly. If current evidence is absent or conflicting and the schema has a matching code, include camera_evidence_missing, food_service_evidence_missing, or late_night_monitoring_needed.

Posture, Monitoring, And Escalation

• deny: uncontrolled major risk, unresolved legal bar, or no viable restricted-issuance basis.
• request_follow_up: material verification gaps, open/referred incidents, unresolved tax issues, or conflicting site evidence.
• issue_restricted: active controls cover the material risks and remaining work is monitorable through ordinary first-90-day checks.
• Build the first-90-day plan from active controls, verification gaps, and open risks. Put critical verification in first_30_days, field observation in days_31_60, and log/review checks in days_61_90, unless the schema requires lexical order.
• Escalation triggers should mirror the unresolved risks and control failure modes, using only allowed trigger enums.

Alcohol Renewal Manual-Review Queues

Fetch target alcohol licensees, alcohol violations, and renewal rules. Use the release boundary in the prompt or the matching renewal rule.

Matching

• Review only current target licenses unless the prompt says otherwise.
• Include violations with violation_date on or before the boundary date.
• Exclude post-boundary rows from queue calculations and list their IDs in the summary when requested.
• Prefer exact license_no matches.
• If a current license has successor_to, include predecessor-license violations on or before the boundary and mark the target's match confidence uncertain.
• If exact license is absent but normalized address/facility evidence clearly matches, use close_address when allowed.
• If both exact and predecessor/address matches are used for one target, use the least-certain confidence among them.

Queue Fields

• matched_violation_ids: sort by violation date ascending, then violation ID ascending.
• violation_count: number of matched pre-boundary violations.
• most_recent_violation_date: latest matched pre-boundary violation date.
• match_confidence: exact, close_address, or uncertain, per the evidence used.
• risk_tier:
  • high: any serious violation, active/open/pending serious matter, unresolved unpaid fine, or uncertain successor match with serious/alert evidence.
  • medium: medium severity, multiple alert flags, fine balance above zero, or open/pending non-serious matter.
  • low: only minor/settled/paid/warning history with no balance or alert concern.
• next_step_label priority:
  • board_review for high-risk serious/open/pending matters.
  • manual_fine_check for any positive fine balance or unpaid-fine theme.
  • manual_ALERT_check for alert-flagged records without a stronger next step.
  • additional_record_check for close/uncertain matches or weak evidence without a stronger next step.

Ranking And Summary

Rank queue entries deterministically:

1. Risk tier: high, then medium, then low.
2. Next-step priority: board review, fine check, alert check, additional record check.
3. Violation count descending.
4. Most recent violation date descending.
5. License number ascending.

Use ranks 1..N with no gaps. Summary fields should be calculated from the queue and excluded rows:

• queue_size: number of queue entries.
• boundary_date: exact boundary date used.
• post_boundary_violation_ids_excluded: excluded IDs sorted lexically unless the schema says otherwise.
• close_or_uncertain_match_license_numbers: target license numbers whose match confidence is not exact, sorted.
• board_review_license_numbers: target license numbers whose next step is board_review, sorted.